MinorBench
收藏资源简介:
MinorBench是一个开源的基准数据集,由新加坡政府科技局创建,旨在评估大型语言模型在拒绝儿童提出的不安全或不适当的查询方面的能力。该数据集的具体内容、大小等详细信息未在文中明确描述,但提及了它包含了儿童可能会向聊天机器人提出的问题,这些问题涉及危险行为、性内容、脏话、仇恨言论、自残和物质使用等风险类别。
MinorBench is an open-source benchmark dataset created by Singapore's Agency for Science, Technology and Research (A*STAR), designed to assess the ability of large language models (LLMs) to refuse unsafe or inappropriate queries raised by children. Detailed information such as the specific content and scale of this dataset is not explicitly specified in the source text, but it is noted that the dataset includes questions that children may ask chatbots, covering risk categories including dangerous behaviors, sexually explicit content, profanity, hate speech, self-harm, and substance use, among others.




